A MTMCMC-Based Dark Spot Detection and Feature Extraction Under the Energy Function Framework

• Published in: IEEE access 2024-01, Vol.12, p.1-1
• Main Authors: Li, M M., Wei, S W., Wang, Y.
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Dark spot detection; Energy function framework; Energy management; Feature extraction; Fields (mathematics); Image segmentation; Labels; Markov chains; Markov random fields; Monte Carlo methods; MTMCMC algorithm; Oil pollution; Oil spills; Radar detection; Radar imaging; Radar polarimetry; Sentinel-1; Speckle; Synthetic aperture radar; Tessellation; Training
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3368128

Dark spot detection is an important and fundamental step in oil spill detection, which affects the accuracy of oil spill detection. In this paper, a Multiple-Try Markov Chain Monte Carlo (MTMCMC)-based dark spot detection method is presented under the energy function framework. Firstly, subregions, which are partitioned by geometric tessellation, replace pixels as detection units to overcome the influence of speckle noise on dark spot detection of Synthetic Aperture Radar (SAR) image. Based on the partitioned subregions, the model of characteristic random field is defined by K-S distance, and the model of label random field is established by Markov Random Field and improved Potts model. A dark spot detection model based on the partitioned subregions, that is, a regionalized dark spot detection model, is established by combining the models of characteristic random field and label random field under the energy function framework. Further, a MTMCMC algorithm is applied to simulate from the dark spot detection model to identify dark spots; In the MTMCMC algorithm, multiple try strategy is realized by defining the weight probabilities according to the detection model and proposal distribution. The proposed and three comparison methods are tested on Sentinel-1A SAR images. The average Producer's, User's and Overall accuracies of the proposed method are 97.5%, 97.9% and 98.3%, and its average Kappa coefficient is up to 0.954. Compared with the test results of three comparison methods, it can be found the defined weight probabilities of the MTMCMC can select more suitable candidate parameters of the dark spot detection model to improve the detection accuracy, and the proposed method can detect dark spots more precisely. Finally, dark spot detection results by the proposed method are used to extract the statistic and geometric features.